1. Technical Field of the Invention
The present invention relates to a system for changing an attribute of image by electronic watermarks, wherein the attribute of image with an inserted watermark is changed by adding another electronic watermark.
2. Description of the Prior Art
Recently, it has become important to prevent reproduction of image information. Therefore, Image information are accompanied very often by some kind of attributes which are identifiers to instruct commands such as xe2x80x9creproduction prohibitedxe2x80x9d, or xe2x80x9creproduction allowed only oncexe2x80x9d.
One of the conventional techniques for adding the attribute to image information is disclosed, for example, in JP 08275127 A, 1996 (Video signal processing apparatus). Referring to this document, video identifier, or video blanking identifier (VBI) is inserted in vertical blanking periods to control the reproduction of image information. VBIs are inserted in the 20-th vertical blanking period of the first field and 283-th vertical blanking period of the second field. Further, in effective video signal areas after horizontal synchronizing signals and color burst signals, 2 bit reference signal and 20 bit information signal are inserted. The two bits from the above-mentioned 20 bit information signal are used for copy generation management system (CGMS) which controls the reproduction of image information.
VBI and CGMS are applicable for analogue video signals in 525 line NTSC, 625 line PAL, and 1125 line High Definition Video System.
However, VBI can be easily eliminated by filtering, because VBI are inserted in fixed positions. Therefore, illegal reproduction can not be prevented perfectly.
Accordingly, visible or invisible electronic watermark becomes an alternative.
Here, the visible electronic watermark has an effect to appeal visually not to reproduce and misappropriate the displayed image information, because prescribed charactes and figures are added to image information by using the visible electronic watermark.
The visible electronic watermark is disclosed, for example, in JP 08241403 A, 1996 (Digital watermarking without color change in image). Referring to this document, only brightness is changed in picture elemnts corresponding to opaque part of the visible electronic watermark data. The brightness is determined by such parameters as color components, random numbers, and values of picture elements of the visible watermark data.
On the other hand, the invisible electronic watermark is recognized visually, because it is imbedded in image information without degrading the image quality.
For example, author of image information may be identified by the invisible electronic watermark. Further, the invisible electronic watermark may notify users that the image information is prohibited to reproduce, or actuate reproduction prohibition mechanism in VTR.
The invisible electronic watermark may be inserted, for example, in the least significant bit (LSB) in picture element, because the LSB does not affect so much the image quality. However, the LSB is eliminated easily by low pass filtering, or image information compression. Therefore, in this case, it becomes difficult to detect filtered or compressed invisible electronic watermarks.
Another electronic watermark is disclosed in Nikkei Electronics 1996, 4.22, No. 660, 13 p). Referring to this document, Image information is transformed in frequency space, and the electronic watermark is imbedded in the dominant frequency components. Therefore, the electronic watermark is not easily eliminated by filtering or image compression. Furthermore, when random numbers, for example, under the normal distribution, are employed, the electronic watermarks rarely interfere with each other, do not affect image quality so much, and are not easily destroyed.
Here, the electronic watermark by using the frequency transformation is explained, referring to FIGS. 7 to 11.
As shown in FIG. 7, original image 701 is transformed by discrete cosine transformation (DCT) 703. xe2x80x9cnxe2x80x9d data f(1),f(2), . . . ,f(n) is chosen among the frequency components of which amplitudes are great. Further, an electronic watermark of which frequency components are w(1),w(2), . . . ,w(n) is assumed to follow the normal distribution with average 0 and variance 1. Watermark imbedding means 704 calculates for each xe2x80x9cixe2x80x9d
F(i)=f(i)+xcex1xc2x7|f(i)|xc2x7w(i), 
where xcex1 is a scaling factor
Then, output image 710 with watermark is obtained through inverse DCT means 709.
The imbedded watermark is detected on the basis of known original image data and known w(i) where i=1 to n.
As shown in FIG. 8, object image 802 with w(i) is transformed by DCT 804 to obtain its frequency components F(1), F(2), . . . ,F(N). At the same time, original image 801 is transformed by DCT 803 to obtain its frequency components f(1), f(2), . . . ,f(N). Then, the watermark data W(i) is calculated. Here, W(i) is defined as (F(i)xe2x88x92f(i))/f(i). Further, statistical similarity C is defined as (Wxc2x7w)/(WDxc2x7wD), where (Wxc2x7w) is a scalar product, WD and wD are the absolute values of W and w, respectively. Finally, statistical similarity determination means 810 determines that electronic watermark is imbedded in object image 802, when C is greater than a prescribed value.
The watermark detection by the statistical similarity requires the original image. Therefore, the author of the original image can use this method. However, general users of playback apparatus such as, for example, VTR, can not use this method. Accordingly, the following method is proposed for the general users of MPEG (Moving Picture Experts Group) system which handles compressed moving pictures, according to ISO (International Standardization Organization) protocol.
As shown in FIG. 9, in the above-mentioned method for MPEG system, electronic watermark is embedded in blocks which are the areas of 8 pixelsxc3x978 pixels divided from the original image.
At first, the original image is transformed by DCT 903 to obtain f(1),f(2), . . . ,f(n) where f(1),f(2), . . . ,f(n) are selected in order of amplitude of the obtained frequency components and f(1) is minimum. Again, electronic watermark is assumed to follow the normal distribution with average 0 and variance 1. Then, electronic watermark embedding means 905 calculates for each xe2x80x9cixe2x80x9d
F(i)=f(i)+xcex1xc2x7avg(f(i))xc2x7w(i) 
Here, xcex1 is a scaling factor, and avg(f (i)) is an average of |f (ixe2x88x921)|, |f (i)|, and |f (i+1)|. In the following, avg(f (i)) is referred to as partial average. Then, MPEG data is obtained through quantization and coding.
The original image data is not required. Only the data w(i) is required, where xe2x80x9cixe2x80x9d is 1 to n.
As shown in FIG. 10, after inverse quantization of MPEG data, F(1),F(2), . . . ,F(n) are selected in order of amplitude of the obtained frequency components, where F(1) is minimum. Then, electronic watermark data W(i) defined as F(i)/avg(F(i)) is calculated for each xe2x80x9cixe2x80x9d. Further, WF(i) defined as a summation of W(i) over one frame (one picture) of the expanded MPEG image 1101 is calculated for each xe2x80x9cixe2x80x9d. Next, statistical similarity C defined as (WFxc2x7w)/(WFDxc2x7wD), where (WFxc2x7w) is a scalar product, WFD and wD are the absolute values of WF and w, respectively. Finally, statistical similarity determination means 1109 determines that electronic watermark is imbedded in the expanded MPEG data, when C is greater than a prescribed value.
As shown in FIG. 11, electronic watermark is also detected from expanded MPEG image. After DCT of expanded MPEG data 1101, F(1),F(2), . . . ,F(n) are selected in order of amplitude of the obtained frequency components of image data 1101, where F(1) is minimum. Then, electronic watermark data W(i) defined as F(i)/avg(F(i)) is calculated for each xe2x80x9cixe2x80x9d. Further, WF(i) defined as a summation of W(i) over one frame (one picture) of the original image is calculated for each xe2x80x9cixe2x80x9d. Next, statistical similarity C defined as (WFxc2x7w)/(WFDxc2x7wD), where (WFxc2x7w) is a scalar product, WFD and wD are the absolute values of WF and w, respectively. Finally, statistical similarity determination means 1109 determines that electronic watermark is imbedded in othe expanded MPEG image 1101, when C is greater than a prescribed value.
However, a scale of circuit for changing the attribute of image becomes too big, if the same procedure is executed twice, when the attribute of image transformed by DCT should be changed.
Further, the quality of output image is degraded, if the same procedure is executed twice. This is because DCT is executed for both of the detection of attribute and the change of attribute.
Therefore, an object of the present invention is to change the attribute of input image, after detecting the attribute of the input image by executing DCT.
A system for changing attribute of image by electronic watermark of the present invention adds a prescribed insert data to an electronic watermark embedded in an image.
The attribute change system of the present invention comprises: an image input means for inputting image data in which electronic watermark is imbedded; a discrete cosine transformation (DCT) means for transforming the image data; a determination means for calculating a statistical similarity between the electronic watermark and a prescribed electronic watermark; an attribute detection means for outputting an attribute of the image data on the basis of an output from the determination means; a plurality of memories for storing said insert data; a selector for selecting one of the insert data; an insert data register for storing an output from said selector; a brightness register for storing brightness signal of the image data; a color difference register for storing color difference signal of the image data; and an adder for adding the output of the brightness register and the output of the color difference register.
According to the present invention, both the inputted image and the output image possess electronic watermarks, because the system of the present invention can change the attribute of image. Therefore, illegal reproduction of image can be prevented by providing apparates of general public such as VTR with the system of the present invention.
Further, the attribute is not always changed, because the system of the present invention is provided with selector. Furthermore, output image is obtained on the real time basis, because the insert data is inserted into the inputted image successively.
Further, electronic watermark can be inserted and can be changed in case of analog composite signal.